14 May 2026 · 6 min read
Most plants we walk into already own more maintenance data than they use. The gap is rarely sensors — it is asset registers that are out of date, work orders closed without a failure code, and spares that take weeks to land. Predictive maintenance works, but it is the last step in a sequence, not the first.
Start with the failure history you already have
Before buying vibration sensors, pull twelve months of unplanned stoppages and sort them by lost production hours rather than by frequency. In most operations a small number of assets account for the majority of lost output, and those are the only assets worth instrumenting first.
If stoppage records are thin, a fortnight of disciplined logging — asset, start time, restart time, suspected cause — gives you enough to prioritise. This costs nothing and prevents spending on monitoring for equipment that was never the constraint.
Match the technique to the failure mode
Condition monitoring is not one technology. Each method detects a specific class of developing fault, and choosing the wrong one produces data nobody acts on.
- Vibration analysis — bearing wear, misalignment, imbalance and looseness on rotating equipment such as pumps, fans and gearboxes.
- Thermography — loose or corroded electrical terminations, unbalanced phases and overloaded circuits in switchgear and MCCs.
- Motor current signature analysis — rotor bar and winding problems, detectable from the MCC without opening the machine.
- Oil analysis — contamination and wear metals in gearboxes, compressors and hydraulic systems.
- Ultrasound — compressed air and steam leaks, plus early-stage bearing lubrication faults.
Expose the data your control system already produces
Modern PLCs and drives log run hours, trip codes, current draw and temperature. On many older lines this data exists but terminates at the panel door. Bringing it onto a plant network — a historian, or even a well-built SCADA trend screen — is often cheaper than a sensor programme and delivers earlier warning of drifting process conditions.
Where a retrofit is required, it is usually worth handling it during a planned shutdown alongside switchgear inspection, so a single outage covers both.
Predictive maintenance fails without spares availability
A prediction is only useful if you can act on it inside the warning window. If a critical bearing carries a ten-week import lead time, a four-week alert changes nothing except how long you know the outage is coming.
This is why we treat condition monitoring and procurement strategy as one exercise: identify the critical spares behind your monitored assets, agree a min/max holding for them, and put a framework arrangement in place so replenishment does not restart the sourcing process each time.
Power quality is the hidden variable
In operations running on a mix of grid supply and generation, a meaningful share of 'mechanical' failures trace back to voltage dips, phase imbalance and harmonic distortion shortening motor and drive life. Logging power quality at the intake for a few weeks frequently reshapes the maintenance plan more than any sensor.
Sequence the work: rank assets by lost output, fix the failure-reporting discipline, expose the data your controls already generate, secure the spares behind critical assets, then add condition monitoring where a fault develops slowly enough to act on.
If any of this maps to a live decision on your side, our engineering and procurement teams are glad to help you frame it.